A MATLAB toolbox for training and implementing physics-guided neural network-based feedforward controllers

M. Bolderman; Mircea Lazar; Hans Butler

doi:10.1016/j.ifacol.2023.10.1732

A MATLAB toolbox for training and implementing physics-guided neural network-based feedforward controllers

M. Bolderman (Corresponding author)
, Mircea Lazar (Corresponding author)
, Hans Butler (Corresponding author)

Research output: Contribution to journal › Conference article › peer-review

1 Citation (Scopus)

287 Downloads (Pure)

Abstract

Physics-guided neural networks (PGNNs) enable accurate identification of inverse system dynamics by effectively embedding a known physical model within a neural network (NN), and thereby achieve high performance when implemented as feedforward controllers. However, training PGNNs using existing NN toolboxes is complicated. Therefore, this paper presents a MATLAB toolbox that systematically implements, trains, and validates PGNNs. Dedicated functions implement recent results that have been proposed in literature, i.e., we ensure that the PGNN converges to a value of the cost function that is strictly upperbounded by the value obtained when using only the physical model, while also imposing a form of graceful degradation when the trained PGNN is used on data that was not present in the training data. The toolbox is available at:https://github.com/mbolderman/PGNN-Toolbox/

Original language	English
Pages (from-to)	4068-4073
Number of pages	6
Journal	IFAC-PapersOnLine
Volume	56
Issue number	2
DOIs	https://doi.org/10.1016/j.ifacol.2023.10.1732
Publication status	Published - 1 Jul 2023
Event	22nd World Congress of the International Federation of Automatic Control (IFAC 2023 World Congress) - Yokohama, Japan Duration: 9 Jul 2023 → 14 Jul 2023 Conference number: 22 https://www.ifac2023.org/

Funding

This work is part of the research programme with project number 17973, which is (partly) financed by the Dutch Research Council (NWO).

Funders
Nederlandse Organisatie voor Wetenschappelijk Onderzoek

Keywords

Mechatronics
data-based control
identification for control
motion control systems
software for system identification

Access to Document

10.1016/j.ifacol.2023.10.1732

pdfFinal published version, 1.24 MBLicence: CC BY-NC-ND

Cite this

@article{87cda1f0c34c4ce88ca8ab6adf03f45b,

title = "A MATLAB toolbox for training and implementing physics-guided neural network-based feedforward controllers",

abstract = "Physics-guided neural networks (PGNNs) enable accurate identification of inverse system dynamics by effectively embedding a known physical model within a neural network (NN), and thereby achieve high performance when implemented as feedforward controllers. However, training PGNNs using existing NN toolboxes is complicated. Therefore, this paper presents a MATLAB toolbox that systematically implements, trains, and validates PGNNs. Dedicated functions implement recent results that have been proposed in literature, i.e., we ensure that the PGNN converges to a value of the cost function that is strictly upperbounded by the value obtained when using only the physical model, while also imposing a form of graceful degradation when the trained PGNN is used on data that was not present in the training data. The toolbox is available at:https://github.com/mbolderman/PGNN-Toolbox/",

keywords = "Mechatronics, data-based control, identification for control, motion control systems, software for system identification",

author = "M. Bolderman and Mircea Lazar and Hans Butler",

year = "2023",

month = jul,

day = "1",

doi = "10.1016/j.ifacol.2023.10.1732",

language = "English",

volume = "56",

pages = "4068--4073",

journal = "IFAC-PapersOnLine",

issn = "2405-8963",

publisher = "Elsevier",

number = "2",

note = "22nd World Congress of the International Federation of Automatic Control (IFAC 2023 World Congress) ; Conference date: 09-07-2023 Through 14-07-2023",

url = "https://www.ifac2023.org/",

}

TY - JOUR

T1 - A MATLAB toolbox for training and implementing physics-guided neural network-based feedforward controllers

AU - Bolderman, M.

AU - Lazar, Mircea

AU - Butler, Hans

N1 - Conference code: 22

PY - 2023/7/1

Y1 - 2023/7/1

N2 - Physics-guided neural networks (PGNNs) enable accurate identification of inverse system dynamics by effectively embedding a known physical model within a neural network (NN), and thereby achieve high performance when implemented as feedforward controllers. However, training PGNNs using existing NN toolboxes is complicated. Therefore, this paper presents a MATLAB toolbox that systematically implements, trains, and validates PGNNs. Dedicated functions implement recent results that have been proposed in literature, i.e., we ensure that the PGNN converges to a value of the cost function that is strictly upperbounded by the value obtained when using only the physical model, while also imposing a form of graceful degradation when the trained PGNN is used on data that was not present in the training data. The toolbox is available at:https://github.com/mbolderman/PGNN-Toolbox/

AB - Physics-guided neural networks (PGNNs) enable accurate identification of inverse system dynamics by effectively embedding a known physical model within a neural network (NN), and thereby achieve high performance when implemented as feedforward controllers. However, training PGNNs using existing NN toolboxes is complicated. Therefore, this paper presents a MATLAB toolbox that systematically implements, trains, and validates PGNNs. Dedicated functions implement recent results that have been proposed in literature, i.e., we ensure that the PGNN converges to a value of the cost function that is strictly upperbounded by the value obtained when using only the physical model, while also imposing a form of graceful degradation when the trained PGNN is used on data that was not present in the training data. The toolbox is available at:https://github.com/mbolderman/PGNN-Toolbox/

KW - Mechatronics

KW - data-based control

KW - identification for control

KW - motion control systems

KW - software for system identification

UR - https://www.scopus.com/pages/publications/85184963009

U2 - 10.1016/j.ifacol.2023.10.1732

DO - 10.1016/j.ifacol.2023.10.1732

M3 - Conference article

SN - 2405-8963

VL - 56

SP - 4068

EP - 4073

JO - IFAC-PapersOnLine

JF - IFAC-PapersOnLine

IS - 2

T2 - 22nd World Congress of the International Federation of Automatic Control (IFAC 2023 World Congress)

Y2 - 9 July 2023 through 14 July 2023

ER -

A MATLAB toolbox for training and implementing physics-guided neural network-based feedforward controllers

Abstract

Funding

Keywords

Access to Document

Other files and links

Fingerprint

Cite this